"Small form factor" modular multichannel TPA3255 amp build

[MCH]

Hi,
I was hesitating posting this project, as i guess there will not be much interest here, but questions start to accumulate and I thought that maybe someone wants to have a look and chime in.

You might recall me asking about using heat pipes in audio amplifiers. What i had in mind was this project, and after doubting for a while i decided to give it a go, at least until i find a technical road block that i cannot solve.

The objectives of the project are the following:
1. As small as possible, low cost TPA3255 multichannel amplifier.
2. Modular: it will consist on small stereo boards that can be stacked in pairs, as as many as needed. (I will probably print 5 PCBs and stick to 8 channels with 1 PCB extra in case)
3. Cooling is via heat pipes/radiator/fan (this was the original idea of the project and i will stick to it)
4. Cooling is enough to use a 3D printed enclosure: This is for me an important part of the project, as it will allow to build without any enclosure restriction. Here i might cheat a bit. I have access to polyimide in liquid form that has a Tg above 250°C and can resist up to 400°C. Obviously i cannot 3D print this, but i might prepare some protection patches for the critical parts if needed (i.e. to hold the radiator).

I based the design on TI evaluation board with the following modifications:
- Changed to Würth 7443631000 inductors because they are cheaper, more available to me, and according to TI perform same or better than the coilcraft toroidals in the evaluation board.
- Changed the 4700uF bulk capacitors for smaller and cheaper 18x30 2200 uF laying horizontal a la Aiyima/Fosi. Not sure about the consequences of this.
- Stripped the board of everything that I don't need and everything that can be centralized in a "power and control board":
- All the voltage regulation (48V to 15V, 12V and 3.3V will be centralized and shared between the different stereo boards)
- The amp boards will have the option to operate in BTL and PBTL (mono) modes, but not SE mode.
- All the logic, error messages, reset etc will take place in the control board
- Added the following:
- Added PFFB circuit and changed the necessary values of the original circuit to addapt to PFFB operation
- Added centralized audio sense circuit (will be possible to defeat)
- The amp boards will not have connectors. All power and signals will be soldered cable to board. This will allow to save a lot of space and budget. Of course, the other end of the cables will have connectors to connect to speakers and control board

As of now, the amp boards measure 86x66 mm. The control board measures 97x50 mm and can accommodate 4 amp boards.

 

[MCH]

Something else that i needed to change from the Evaluation board is obviously the power and output traces, to adapt to the smallest form factor:

The top picture shows the main power trace in the center and some of the output traces to the right. The lower picture shows the 4 output traces as they get out of the chip towards the inductors. As you can see, i neede to leave the two central traces narrower than the external ones. These narrow traces are still 3.5 mm wide, and according to digikey trace calculator they would still be able to pass 6 amps with a temperature increase of 10°C only. I wonder how reliable these calculators are.

Anyways, my question here is more related to the separation between traces. I have notice TI leaves a wider separation between output traces that belong to different channels and very narrow separation between power and output traces and between +/- output signals for the same channel. I am doing the same in my design, never going below the separation in TI design. However, in TI design, the traces divert early on, while in mine they run in parallel for quite a while. Do you guys think this is going to be a problem?

Left: TI design where signal traces start very close but divert early on. Right my design where signal traces remain parallel for a while, but with R and L channel well separated. in my design, the separation between channels is 1.47 mm and between +/- for the same channel is 0.45 mm.

 

[MCH]

The other important power trace is in the auxiliar "control board".
This will be the trace subject to the largest current, as it will divert the power supply to the 4 different amp boards. Here i do include connectors. I will be using JST B02P-VL, that are the same that some Hypex PSU use and are specified for up to 20 A. The trace in question that connects all outputs to the input is 7.1 mm wide and i placed it on both top and bottom side of the PCB. I don't know if this is a good practice. I also thought that, if necessary, i would leave this trace naked and plate it with tin on top. As is, according to the trace calculator, each trace can carry 10 Amps within a 10°C. Again, it seems too good to be true to me.

The trace in question can be seen between the blue power connectors:

seen from the bottom side (all other traces not shown):

 

[mcdn]

Excellent stuff! As a big fan of the TPA3255 I love to see this

For the race widths you could always go for 2oz copper if you want to play it really safe or even make things more compact. I suppose at a maximum 100W continuous for 8 channels you'll need 16A in that power distribution bus, so heavier copper could help there, but is unnecessary in the individual modules.

I wouldn't be worried personally about the parallel traces either, the TI ref design just happens to be laid out that way.

 

[MCH]

Excellent stuff! As a big fan of the TPA3255 I love to see this

For the race widths you could always go for 2oz copper if you want to play it really safe or even make things more compact. I suppose at a maximum 100W continuous for 8 channels you'll need 16A in that power distribution bus, so heavier copper could help there, but is unnecessary in the individual modules.

I wouldn't be worried personally about the parallel traces either, the TI ref design just happens to be laid out that way.

thanks mcdn,
Actually, the evaluation board is all 2oz copper (that might be an overkill). I was just thinking that if what the trace calculators say is half true, i would be good even with 1oz. You are not a fan of over plating that single bus then? the only disadvantage i see is that if one day i need to rework something (quite likely) it is going to be a mess.... I just checked jlcpcb and the difference 1oz vs 2oz is 16 euros, ouch!

Anyways, two additional questions, about sharing the regulators between the amp boards:
- Both the buck regulator (48 -> 15 V) and the LDO (15V -> 12V) are rated 1A. I am assuming that this is enough for powering all the op amps (2 per board) and TPA3255. I understand that all these components draw very little current. (TPA3255 at 50% duty cycle uses 74mA of the 12V supply according to datasheet). Am i missing something?
- 48V filter caps: the evaluation board has the following filter caps right at the input of the 48V supply:

I placed this exact combination at the 48V input of the shared board, before the regulators.
But C39 is quite bulky and i want to avoid placing the same for each individual amp board. What i temporarly did is to add C3 and C11 to each amp board, but left C39 out. I suspect that i am good like that, and adding the same 3 caps filter in the shared board and then to each individual amp board is an overkill. I guess until i dont measure i dont know, but what do you guys think is reasonable to do?

PS: @mcdn as you can see i went for the analog audio sense circuit. For the moment i want to try and leave the microcontroller separate taking care only of measuring temperatures and control the fan. Don't want to risk all the project for a bad microcontroller programm. But this might change later on.

 

[MCH]

Today Amir published the review of the new Fosi V3 mono TPA3255.
Since i am working in this project. i can't avoid having a look at the parts that these commercial amps use, and most specifically to the output LC filter. What interests me is finding parts that are smaller than the ones specified by TI in the evaluation board. As you can see in my design, output capacitors and inductors alone account for ca. 50% of the space, and for a reason!:

TI describes with great detail how to select the output filter components, and for the capacitors, the say the following:

So for the filter capacitors, the important parameter is dV/dt, and the recommendation is to be well above 23 V/us, ideally 100 V/us. And I tell you something, the film capacitors with high dV/dt have one thing in common, they are BIG
Indeed you can see the size of the Kemet capacitors in the evaluation board and in my design, with a healthy 120 V/us

But where do Fosi Aiyima & co. find these small capacitors suitable for a high power TPA3255 design?? I have been checking dozens of datasheets and all these small caps have dV/dt around 15 V/us....

Fosi released pictures of the inside of the new V3, and claims that the caps are Wima:

from the picture it is easy to deduct that the caps must be 7.2x7.2 mm (compare the opamp that measures 10x6mm or the large electrolitics, that have a diameter of 15 mm and a pin separation of 7.5mm)
Well, i promise i have checked all the datasheets of 1uF 100V 7.2x7.2mm Wima caps in Mouser (and there are like 30), and they all have dv/dt way lower than recommended

Example MKS2D041001K00JI00:

Guys, this is not mean to be any conspiracy theory, i am sure Fosi knows what they do and i would buy their amp tomorrow if i had a use for it, but someone please help me to identify these caps, they are less than half the size of the ones i plan to use, and i want them!!!!

 

[mcdn]

Unlikely as it seems, I think the TI note overestimates the required slew rate. 50*2*pi*73/1000 is indeed 23, but for slew rate I thought the peak amplitude was supposed to be used, so 25V not 50V, which gives 11.5V/uS. Not a great margin for error but not actually outside spec. It also looks from Amir's graphs like they've chosen a lower filter bandwidth of around 50kHz, and the supply voltage is 48 not 50, which would give 24*2*pi*50/1000 = 7.5.

 

[kemmler3D]

This is pretty fascinating - I have no electronics knowledge so can't help, but I'm really interested in what this amp will be used for!

 

[MCH]

Unlikely as it seems, I think the TI note overestimates the required slew rate. 50*2*pi*73/1000 is indeed 23, but for slew rate I thought the peak amplitude was supposed to be used, so 25V not 50V, which gives 11.5V/uS. Not a great margin for error but not actually outside spec. It also looks from Amir's graphs like they've chosen a lower filter bandwidth of around 50kHz, and the supply voltage is 48 not 50, which would give 24*2*pi*50/1000 = 7.5.

According to TI you need to take the vias (half of the power supply voltage) into account:

with this:

with their 50V example it just happens that they assume a 50V power supply as well so the result is the same, but when i did the calculation with the max. power at 4Ohm from Amir's review and a 48V power supply i got 20 V/us at 73kHz, that would still be 13.7 V/us at 50 kHz,

But on top of that you still need to count the overshot at the edge of the square wave, that even if with PFFB can be optimized, still can account for a good percentage upwards according to TI examples (these graphs are after optimization):

without PFFB or without load it goes out of the roof. And i believe that is the reason why TI recommends minimum 100V/us even if their 50V example is just 23V/us.
In any case, probably 100V/us is conservative, but 15V/us seems to me unecessarily underspecified, specially considering they have plenty of PCB real state in the mono version and the next size of Wima caps (11x8mm) are specified for 70V/us, that would probably be a wiser choice. BTW, might switch to those 11x8 wima myself....

 

[restorer-john]

Centralised audio sense circuit.

Are you monitoring all 8 inputs? Adjustable trigger levels and adj timeout? Bandwidth, can it be triggered by HF noise from D/As PSUs or even video signals (which can be useful). Muting clicks and DC offset from the source etc.

 

[MCH]

Thanks to @daniboun (thanks man!) i could identify the filter caps in the 3E audio TPA3255 boards, and it seems they check all the boxes, including price (0.8 euros) and availability:

--> TDK EPCOS B32529D1105J

1uF 5% 100V 300 V/us (!!) 7.8 x 7.8 mm (vs Kemet 18 x 9.5 mm)

The only spec that is, obviously, much inferior to the Kemet is the derrating at high frequencies. But i don't think it is something to worry about (I hope):

(we are looking at the 1.0 uF curve)

These would allow me to reduce significantly the size of the PCB in the vertical direction, now the question is, do i want to redo all the parts placement?? grrrrrr

 

[daniboun]

Thanks to @daniboun (thanks man!) i could identify the filter caps in the 3E audio TPA3255 boards, and it seems they check all the boxes, including price (0.8 euros) and availability:

View attachment 361580 --> TDK EPCOS B32529D1105J

1uF 5% 100V 300 V/us (!!) 7.8 x 7.8 mm (vs Kemet 18 x 9.5 mm)

The only spec that is, obviously, much inferior to the Kemet is the derrating at high frequencies. But i don't think it is something to worry about (I hope):
View attachment 361583
(we are looking at the 1.0 uF curve)

These would allow me to reduce significantly the size of the PCB in the vertical direction, now the question is, do i want to redo all the parts placement?? grrrrrr

View attachment 361585

I would say that it is above all a project in its own right... passion has no limits))


This could also be interesting for you :

Inductors testing

 

[MCH]

I would say that it is above all a project in its own right... passion has no limits))


This could also be interesting for you :

View attachment 361587
View attachment 361588

Yes, that is in the same LC filter design document that also specifies the caps. For inductors I went for the 10uH Würth, that perform well, and are cheaper and more available than the coilcraft.

 

[MCH]

Hello guys,
Was placing the traces in what i thought was the final version, the smaller caps allowed me to reduce the amp board a couple of mm more but most importantly will allow me to place a second heat pipe on the area where the power input and output traces are. I don't think it will be necessary but it is good i will have the space to place one there if needed.

But i realized i left one detail unattended...

The power and control board is aimed to get power from one single power supply and split it to up to 4 amp boards. They are just connected through a wide track.
I was thinking that each amp board has 2x2200 uF caps connected straight to power. With 4 boards i will have 8x2200 uF.... and that seems quite some in-rush current to me....

I think i have two options:
- go for smaller caps - TI recommends at least 1000 uF, what would reduce the total capacitance to 8000 uF, but still....
- keep the 2200 uF caps and use some sort of sequential switch. And here is my question:

It seems that the obvious solution for a sequential switch would be a timer IC or microcontroller and a mechanical relay per amp. However, i would like, if possible, to avoid relays for different reasons (relays are bulky or expensive, i prefer not having several clicks sequentially every time the amp turns on, I understand solid state relays are not really suitable for DC....).
I wanted to ask if anyone knows about any sequential switch in the form of IC. The specs should be:
- At least 3 channels
- At least 50V max
- At least 3A per channel, ideally 5A
I have searched Mouser & co. but everything i find lacks something. Examples:
- Texas Instruments TPS1HTC30 : voltage and current are fine, but only one channel. Would have to have 3-4 of them activated externally. This could be a substitute for the relays.
- ST IPS4260L: 4 channels, but 0.5A max and i believe the 4 channels switch on simultaneously

Does something similar to these IC exist that has more than one channel that switch on sequentially or am I asking too much?
Am I worrying too much and the in-rush current with 8x2200uF won't be an issue?
would a simple thermistor do the job? better one per board? i dont really like the idea of having a thermistor there always hot and eating voltage, but if there is no other alternative...
Comments welcome. Thanks!

 

[MCH]

Actually, I am liking the TI TPS1HTC30, as it has built in Current Limit functionality that makes the relays or delayed switches moot.

It is a simple high voltage (up to 60V) high current (up to 16A) switch via a logic EN pin. The way it works is that it allows you to set a maximum current that passes through the switch depending on the value of an external resistor that connects to a control pin. If that set maximum current is reached, the voltage goes down to avoid going beyond the maximum current set. The cool thing is that you can change the value of that resistor - or just short it to ground - on the fly so that the maximum current changes.

The datasheet even describes the use case of using this current limit functionality to limit the in-rush current while charging capacitors. Exactly what i needed!!

In my case, the sequence could be:
Set the max current to, say, 2A with a resistor -> turn on the PSU -> the caps charge, current is limited to 2A, the voltage increases as the caps get charged and draw less current -> when the voltage reaches the nominal voltage of the PSU, the voltage supervisor (that is already in the circuit to switch on the TPA3255) detects it and the same signal that switches on the TPA3255 can be used to trigger a mosfet that shorts to ground the resistor that limits the current so that the current is not limited anymore.
Genius!

And additionally, besides overcurrent protection, reverse polarity protection, etc, the same IC also has a current mirror that reports how much current is passing (=how much current the amp is using) in the form of voltage, what, if one feels fancy, can be used to build one of these cool power meters (well, it won't report the actual output power but the input power, but nobody has to know this ):

And all this for the IC, 4 resistors, one diode and the mosfet. This is ideal, only wish they had a multi channel version (they do, but with lower max. voltage)

 

[mcdn]

Wouldn't inrush current be handled by the soft-start thermistor on whatever PSU you are using?

 

[MCH]

Thanks, you are probably right. I have no idea about PSUs but I understand the thermistor is normally right at the mains input to avoid in-rush current from the mains to the PSU? (in my case at 240V) Would that be enough to avoid a high current at 40-48V at the amplifiers?

Another case, that I will try to avoid as much as I can, is when one plugs the amplifier to the PSU already plugged to mains. My Aiyimas A07 max make a nice spark when I do that by mistake or laziness.

Actually, what made me think on adding some sort of protection is the thermistor the Aiyimas have at the power input. But I don’t like the idea of having that thing there generating heat all the time, and the TI IC has an on resistance of only 30m Ohm, maybe 40 in my case. It might still get a bit hot but I could also attach a little heat sink to it. What do you think?

 

[mcdn]

I mean technically you don't need separate reservoir caps for each module, you could put one shared set on the control/power board. If you use something like the Connex SMPS800RS you could even just rely on its output caps (2x4700uF) or replace them with bigger ones. Whatever you do, all this sums up to a single capacitative load that the PSU's thermistor is there to control the charge rate of.

 

[MCH]

I mean technically you don't need separate reservoir caps for each module, you could put one shared set on the control/power board. If you use something like the Connex SMPS800RS you could even just rely on its output caps (2x4700uF) or replace them with bigger ones. Whatever you do, all this sums up to a single capacitative load that the PSU's thermistor is there to control the charge rate of.

Sharing bulk capacitance fits perfect with the spirit of this project: simple, small, low cost and modular.
The problem is, TI recommends to place the bulk capacitors as close as possible to the chip, as the heat sink allows....
See drawing and note T1

 

[mcdn]

Of course they do, but I seriously doubt it would make a measurable difference if they were shared in this application.